Neural NOPE: a physics-informed neural network approach to mitigate nonlinearity in pyramid wavefront sensing for adaptive optics

Abstract

Pyramid wavefront sensors (PWFSs) are key components of adaptive optics (AO) systems for next-generation extremely large telescopes (ELTs). However, their nonlinear behavior can degrade image quality under strong atmospheric turbulence, especially when conventional matrix vector multiplication (MVM) methods are used. To address this, we introduce a hybrid approach that combines a model-based reconstructor, called nonlinear pyramid extension (NOPE), with a physics-informed unrolling architecture. Compared to the standard NOPE, the new neural-NOPE accelerates convergence, enhances reconstruction fidelity, and improves robustness while preserving physical interpretability. Comparisons show that neural-NOPE effectively controls PWFS nonlinearity under severe turbulence and outperforms standard MVM approaches in such conditions.